---
title: "Does llms.txt Work? I Checked a Week of Server Logs"
description: "Verified AI crawlers made 764 requests to my site in a week and asked for llms.txt zero times. Here's the server log data and what to do instead."
featured_image: "https://phillipstemann.com/wp-content/uploads/2026/10/llms-txt-cover-en.png"
url: "https://phillipstemann.com/does-llms-txt-work/"
language: "en"
date_published: "2026-10-17T08:00:00+02:00"
date_modified: "2026-10-15T09:24:06+02:00"
---

# Does llms.txt Work? I Checked a Week of Server Logs

llms.txt is on my own GEO checklist. I set it up for clients, and my site has one too. So I finally did what I should have done before recommending it: I opened my server logs and checked who actually reads it.

The answer, after a week of logs and every AI crawler verified against its company’s official IP list: **nobody**. Not one real AI crawler asked for my llms.txt. A lot of them visited the rest of my site.

**The short version:**

- In 7.4 days (8-15 October 2026), verified AI crawlers from OpenAI, Anthropic and Perplexity made **764 requests** to my site.
- **Zero** of them were for /llms.txt.
- The file was requested 11 times in total: by SEO and tech-lookup tools, by three people in a browser, and by me while testing.
- The AI crawlers read normal pages instead: the homepage, my HARO alternatives list, my SEO service page and my Search Console guide.

## What llms.txt is (and isn’t)

llms.txt is a Markdown file at the root of your site that lists your most important pages, so a language model can find the good stuff without wading through navigation and scripts. It was proposed by Jeremy Howard of Answer.AI in September 2024 (here’s the [llms.txt proposal](https://llmstxt.org/)).

It’s a proposal, not a standard. It doesn’t control access: robots.txt still decides which crawlers may fetch what. And none of the big AI companies has publicly committed to reading it. Mine is generated automatically by Rank Math, which is how most WordPress sites get one.

## How I checked

- **The data:** nginx access logs for phillipstemann.com from 8 to 15 October 2026. 69,015 requests in total.
- **Nothing hidden by the CDN:** my site runs behind Cloudflare, which answers some requests from its own cache before they reach my server. I checked the response headers: llms.txt is not cached at Cloudflare (`CODE_OPEN`cf-cache-status: DYNAMIC`CODE_CLOSE`), so every request for it reaches my server and shows up in the logs.
- **Real bots only:** anyone can put “GPTBot” in a user agent, and many scanners do. So I only counted a request as GPTBot, ClaudeBot or PerplexityBot when its IP address was in the list that company publishes. More on that below, because it turned out to matter a lot.

## The result: zero

| Crawler (verified) | What it is | Requests | Asked for llms.txt |
| --- | --- | --- | --- |
| OAI-SearchBot | Builds ChatGPT’s search index | 453 | 0 |
| ClaudeBot | Anthropic’s crawler | 176 | 0 |
| ChatGPT-User | Fetches a page live when someone asks ChatGPT | 56 | 0 |
| PerplexityBot | Builds Perplexity’s index | 51 | 0 |
| GPTBot | Collects training data for OpenAI | 27 | 0 |
| Claude-User | Fetches a page live for a Claude user | 1 | 0 |
| Total |  | 764 | 0 |

************

Google’s, Bing’s and Apple’s crawlers didn’t ask for it either.

So who did? Five of the 11 requests came from SEO and tech-lookup tools (SEOJuice, BuiltWith and Keywords Everywhere), three from people opening the file in a normal browser, and three from me testing it with `curl`. In other words: **SEO tools and SEO people read llms.txt. AI crawlers don’t.**

## What the AI crawlers read instead

They were busy, just not with llms.txt. The pages they fetched most were the homepage, my list of HARO alternatives, my SEO service page, my Google Search Console guide and my SEO podcast list.

The most interesting one is ChatGPT-User. It doesn’t crawl in the background; it fetches a page when a real person asks ChatGPT something and ChatGPT decides to read your page to answer. It fetched 14 of my pages 56 times in that week. That’s ChatGPT reading my content to answer questions, and it went straight to the pages themselves, not to a summary file.

## Caveats, because one week on one site isn’t everything

- **It’s one site.** A big site may see different behaviour. If you have one, check your own logs (I show how below).
- **It’s one week.** A crawler could, in theory, fetch llms.txt once a month. Still, 764 verified requests and not a single one for the file is a clear signal.
- **Things change.** If an AI company starts using llms.txt tomorrow, this data won’t show it. I’ll re-run the check regularly and update this post.

## So should you have an llms.txt?

Keep it if you have one. It’s free, it does no harm, and if you’re on WordPress, Rank Math or Yoast generates it in a click. Just don’t expect it to do anything for your AI visibility right now, and don’t pay anyone to “optimize” it.

What the logs show matters is what AI crawlers actually fetch: your pages. So the work that pays off is the same as good technical SEO:

- Let the crawlers in. Check robots.txt and your CDN’s bot settings, so you don’t block OAI-SearchBot, ChatGPT-User or PerplexityBot by accident.
- Serve real HTML. If your content only appears after JavaScript runs, many AI crawlers won’t see it.
- Write pages that answer specific questions clearly, with facts and numbers a model can quote.
- Get mentioned on other sites. AI assistants build their picture of you from what others say, not just from your own pages.

That’s the work I do in my [GEO service](/services/geo/), and I’ve since moved llms.txt to the “optional” pile on my own checklist.

## How to check your own site

If you have access to your server logs, this takes two minutes:

```
grep "llms.txt" /path/to/access.log
```

If you see GPTBot, ClaudeBot or PerplexityBot in the results, don’t trust the name. Most of the “AI crawlers” in my logs were fakes. Here’s how I checked which AI crawlers were real, including the script I used.